Beruflich Dokumente
Kultur Dokumente
Erkut Sonmez
Sunder Kekre, Nicola Secomandi, Kenneth West
Tepper School of Business, Carnegie Mellon University
Quantitative Models for Production and Communication Networks
Workshop
2009
† Forthcoming at M&SOM
Outline
• Industry Background
• Field Study
Field Study
• Model
• Modeling insights
• Managerial Insights
Managerial Insights
• Conclusions
2
Industry Background
y g
1. Merchant
C stomer
Customer
3 Clearing House
3. Clearing House
2. Financial Institution
4. Financial Institution of Merchant
of Customer Journey of a paper check
3
Decreasing Check Volume
g
ACH
Debit Cards
Credit Cards
Paper Checks
Paper Checks
Check clearing institutions need to adjust their workforce levels via layoffs
Source: Dove Consulting 4
Field Studyy
Major Commercial Bank (CB)
Helped CB to determine new check processing workforce level and
deployment
5
Efficiency / Risk Tradeoff
y
6
Our Contributions
• Modeled tradeoff at the check clearing house
d l d d ff h h k l i h
• Conducted
Conducted model
model‐based
based analysis on real clearing
analysis on real clearing
house data.
• Ch
Check clearing house implemented our
k l i h i l d
recommendations.
7
Noveltyy
• Business problem level
i bl l l
– Consider all relevant costs
– Consider heterogeneity among check
– Model detailed system and cost of risk
– See Gomes and Meile (2002) for review of papers in
check processing industry
• Methodological line of attack
– Use of simulation‐optimization (SO) methodology
coupled with lower bound (LB) analysis on real data
8
Check Clearing House Operations
Check Processing Workflow
Check Processing Workflow Check Types and Deadlines
Check Types and Deadlines
• Check Types :
– On
On‐us
us
Receiving Sorting Check
– Transit Heterogeneity
• Kill
Settlement
• Re‐handle
1%
1%
Reject‐ Dispatch
Encodingg • Different Deadlines:
Repair
– Dispatch Deadline
– Settlement Deadline
9
Data Analysis
Work‐Streams:
Data: check profiles, labor
resources historical fraud patterns
resources, historical fraud patterns. • Different check dollar value,
Different check dollar value,
volume, fraud probability, arrival
time, deadlines
WORKSTREAMS SCALED CHECK VALUE
WS2
WS3
WORKSTREAMS
10
Model
Stochastic Integer Program Simulation‐Optimization
min E[ W
Workforce
f ( ) + Risk C
Cost(X)
C Cost ((X))]
ARENA
X
s.t.
S i Level
Service L l Constraint
C t i ts ( X )
Opt ‐
Quest
11
Modeling Insights
g g
• Quality of OptQuest solution
– Obtain a LB by solving a MIP that ignores workflow
sequencing constraints and uncertainty
– MIP solution (LB) is 6.1% lower than QptQuest
MIP solution (LB) is 6 1% lower than QptQuest
solution in terms total expected cost
• Simulating MIP solution
– 61% higher than OptQuest
g solution in terms of total
expected cost
• Relevance of modeling uncertainty : deterministic
l f d li i d i i i
simulation
– %2
%2.5 higher than stochastic model in terms of total
5 hi h th t h ti d li t ft t l
expected cost 12
Managerial Concerns
g
• What is the sensitivity of expected costs to
h i h ii i f d
workforce level changes?
• What should a good worker‐to‐shift assignment
policy look like?
policy look like?
• What are the benefits of cross‐training?
13
Sensitivity of Expected Cost to Workforce Level
• B
Being
i conservative
i
and having more than
adequate capacity is
TOTAL far better than lacking
capacity.
Workforce Level
14
Worker‐to‐Shift Assignment Policies
Risk Potentials by Shift
15
Worker‐to‐Shift Assignment Policies
g
Number of Workers
Number of Workers Cost
A good assignment policy should consider all risk
A good assignment policy should consider all risk
potentials and check volumes
16
Benefits of Cross‐Training
Total Expected Cost
p
Cost reduction
is 9.83% in
moving from
point A to B
A
B
Cross‐training is most valuable for workforce levels that
Cross‐training is most valuable for workforce levels that
fall in between overload and excess capacity situations 17
Conclusions
• Conducted model‐based analysis at a major commercial bank
• Recommendations implemented
• Managerial Insights:
– Firing too many workers too early can be very costly
– A good worker‐to‐shift assignment policy should consider
two major
j criteria: check volume and risk p potential
– Cross training can lead to significant cost reductions if it is
introduced at the right capacity bands
Erkut Sonmez
Tepper School of Business
esonmez@andrew.cmu.edu
@ d d